Repository navigation
Releases
v0.16.0
Added
- Added session and saved shell command approvals scoped to an exact command and
working directory, with a remote-endpoint disclosure when granting commands
whose output goes to a remote model. - Added
/permissionsin chat to inspect and edit conversation-scoped tool rules,
including exact shell commands and write directories, and explicitly enable or
disable Allow all with a visible status indicator. Changes stayed in the current
conversation; Cancel discarded edits and new conversations restored defaults. - Added
drove chat --dangerously-skip-permissionsto bypass tool confirmations while retaining execution checks
for one run, with a persistent warning (compactUNSAFEin narrow terminals) and no saved permission grants. - Added
make covto report coverage with uncovered lines andmake release-check
to rehearse a semantic-release cycle in a disposable repo with no remote writes. - Chat automatically copied non-empty transcript selections, including rendered Markdown,
code snippets and text spanning messages, while preserving composer drafts, focus
and explicit copy shortcuts.
Changed
- Shell commands trimmed surrounding spaces, tabs and newlines consistently for
approval and execution, and foreground and background tasks ran in the chat's
recorded working directory. - Reused the conversation permission reset when loading saved chats.
- The systemd service now stops a hand-started
drove serveprocess before
starting its own, andmake service-restartwas made to run even when a file
has that name. - Documented unlocked CLI and TTS dependency resolution, with a lock-constrained
reinstall recipe in the existing guides. - Declared directly imported CLI and validation dependencies, including TTS workers,
and removed the unused HTTP/2 dependency extra. - Aligned contributor commit types with the PR title and release parser, with
regression checks covering the release guide too. - Refreshed the contributor guide with current module and engine pointers,
directory-based docs and test inventories, and the distinction between drove
settings and third-party environment conventions; corrected backend comments
to include text-to-speech. - Pull requests now get a
checkstatus that fails on a staleuv.lock, a
ruff error, a mypy error or a failing test, and releases frommainwait for
the same check, so a redmainno longer cuts a version. - The test suite now fails on a never-awaited coroutine warning instead of
printing it; two proxy tests that hid an empty response body this way now
assert on the body.
Fixed
- Fixed an intermittent chat permissions editor test failure on slow machines, which had
blocked a release. - Fixed an intermittent test failure when a timed-out shell command's process
exited during test cleanup, which had blocked a release. - Allowed applying chat permission changes after inspecting an unchanged rule,
without requiring an extra save. - Preserved reviewed changelog bullets in release notes and refreshed the lockfile
with each version bump, and made release publication retryable from the original workflow commit. - Made
service-restartrun even when a file has that name, and stopped
hand-starteddrove serveinstances before starting the systemd service. uv.lockrecords the current drove version again, souv runno longer
dirties the working tree.
v0.15.1
v0.15.1 (2026-11-03)
Bug Fixes
- downloads: Support Xet progress callbacks (
a75b947)
Detailed Changes: v0.15.0...v0.15.1
v0.15.0
v0.15.0 (2026-11-03)
Bug Fixes
-
chat: Align session identities with HTTP URL parsing (
977a1bb) -
chat: Isolate session history and restore conversation state (
3f40c76) -
chat: Preserve drafts and escape literal slash messages (
6517905) -
cli: Handle repeated and oversized watch intervals (
380ba69) -
cli: Reject invalid boolean values and watch intervals (
ccf2082) -
downloads: Write records and sessions through the atomic writer (
492c3ab)
write_state used a fixed temp name that write_text would follow and never cleaned up. The record keeps its umask-default mode and exact bytes. The downloader's link check drops the old temp name: the record's directory is still vetted via the record path, and the temp file relies on exclusive creation.
Refs #156.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- server: Keep cold starts off the manager lock and re-check idle stops (
8153fe8)
A cold start held the global manager lock through the health wait and the prompt-cache restore, so requests for already-loaded models queued behind another model's startup. The idle watcher also checked idleness without the lock and stopped the model after acquiring it without re-checking, so a request claimed in between was cut off.
- Claim a running, up-to-date model without the lock (no await between the check and the claim; every stop unregisters before its first await). - Release the lock while a started backend waits for health and restores its cache; the model stays registered and marked starting, counts toward capacity, is never evicted, and is not claimable until ready. Further requests for it wait and share it. - Skip the prompt-cache save when stopping a model that never became ready. - Re-check under the lock in the idle watcher before stopping. - Make the model-less fallback wait out a start instead of forwarding to an unready backend.
Refs #115
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- server: Never release another task's lock when cancelled mid-relock (
9778ce6)
A start or eviction drain that released the manager lock and was cancelled while queued to take it back left without it, and the enclosing async-with then released the lock held by another task. Shield the re-acquire and re-raise the cancellation once the lock is held; a cancelled start still stops its process and settles its waiters.
Plan review finding PC3 on #115.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- storage: Close the temp descriptor when fdopen fails (
2076e68)
Cleanup only unlinked the path, leaving the exclusively created descriptor open; this shape was copied from the old session and integration writers. Search: grep -rn 'os.fdopen' src -> 2 found, 1 fixed, 1 left
(observe.py:424, the observe record writer, outside #156's scope; deferred).
Refs #156, review B-correctness-1.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
Documentation
Refs #156.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- server: State the lock-free claim contract in ensure_running (
e544351)
The ensure_running and _claim_slot docstrings still said a claim needs the manager lock, though the fast path and request_started now claim without it. Describe the no-await rule and the stop-side safeguards.
Code review finding docs-1 on #115.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
Features
- chat: Keep the message input focused on transcript clicks (
115000b)
Clicking a message focused it and clicking empty transcript space focused the scroll container, so typing stopped reaching the composer. Message and the transcript no longer take focus on click (they stay Tab-reachable for Ctrl+Y / Ctrl+B and keyboard scrolling), and a click on the main chat screen that hits no control returns focus to the input. Modal screens and real controls keep their own focus handling; wheel scrolling and mouse text selection are unaffected.
Closes #140
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
Refactoring
- integrate: Stage agent configs through the atomic writer (
23245d2)
_stage keeps its mkdir and copied permissions; stage, commit, backup and rollback are unchanged.
Refs #156.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- storage: Add one exclusive-create atomic text writer (
c3fffa3)
Readers see the old file or the new one; the temp file has a random, exclusively created name, so a planted file or symlink cannot redirect it, and it is removed on any failure.
Refs #156.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
Detailed Changes: v0.14.0...v0.15.0
v0.14.0
v0.14.0 (2026-10-28)
Bug Fixes
- chat: Align the approval scope line and mark the reason (
2d7df0f)
The #tool-scope label lacked the "message" class its siblings carry, so it started two cells left of the reason line and ran straight into "Arguments:". Give it the class, and give #tool-reason a $warning rule (the project's attention colour) so the reason stands out from the boilerplate intro. Existing tokens only.
Belongs to: #154.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- chat: Cap old tool results when a session is loaded from /sessions (
78f1c32)
Only constructor resume capped stored tool results; choosing an older conversation in /sessions put an oversized result back in history and in the next model request. Cap it at that load too.
Refs review A-correctness-1.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- chat: End a pending mid-fetch approval on stop or quit (
ef258b8)
The worker's wait for a mid-fetch origin approval only noticed a closed loop, so a stop or quit left the dialog open and the turn generating until someone answered. The wait now also watches stop_requested and cancels the approval, which dismisses its dialog and refuses the fetch.
Refs review A-correctness-2.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- chat: Judge NAT64, 6to4 and other IPv4-in-IPv6 forms by the IPv4 (
d3a90dd)
64:ff9b::/96 is globally routable to ipaddress, so a NAT64 address of 169.254.169.254 or 10.0.0.1 passed the internal-address check. Decode IPv4-mapped, IPv4-compatible, NAT64 (well-known and local-use /96 layout), 6to4 and Teredo addresses and apply the IPv4 rule.
Refs review B-security-2.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- chat: Never leave a tool worker waiting on a closed loop (
7e54ab2)
Poll the mid-fetch approval so a worker returns when the event loop is gone, and drop the per-call grant when a call is denied.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- chat: Refuse out-of-root reads before revealing whether they exist (
2a82b01)
read_file, list_files and search_files ran their exists/is_file/is_dir checks before _check_roots, so a call outside the approved roots got "not found" or "not a file" instead of the roots refusal, which leaks whether the path exists. Check the roots first in all three.
Belongs to: #154.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- chat: Scope fetch approvals to scheme, host and port (
07b2fd3)
A fetch grant covered every scheme and port of a host, so approving https://example.com also allowed http://example.com and https://example.com:8443 (another service) without asking, directly or through a redirect. Grants, session approvals, the execution context, redirect checks and internal-address approvals now use the canonical origin scheme://host:port (default port spelled out). Host-only and non-canonical saved grants are ignored, so nothing widens silently.
Docs also note the IPv4-in-IPv6 rule from the previous commit.
Refs review B-security-1.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- chat: Treat local-use NAT64 addresses as internal (
7c36657)
64:ff9b:1::/48 was decoded as if the IPv4 address sat in the last 32 bits, but local-use prefixes may be /48 to /96 with the IPv4 bits elsewhere, so a private target with a public-looking suffix skipped internal approval. Keep decoding the well-known 64:ff9b::/96 and block the whole local-use range unless the user approves the origin.
Refs review B-security-3.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- chat: Warn when an allow-always approval cannot be saved (
5b64ddb)
The persistence failure was swallowed silently. Since #148's set_config_value validates the whole file, any invalid key elsewhere now blocks the save, so the user chose "Allow always", saw nothing, and was asked again next launch. Show a warning notification; the session grant still applies.
Belongs to: #154 (made more likely by #148).
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- cli: Keep logging off the terminal while a Textual app runs (
aa1d19e)
The root callback's basicConfig binds a StreamHandler to the real stderr before any command runs. Textual redirects sys.stderr while it owns the screen, but that handler keeps the original stream, so httpx's INFO "HTTP Request" lines were drawn over drove chat next to the input box.
Swap the root handlers for a single TextualHandler (no stdout/stderr fallback) around every Textual run — chat, the initial model picker, observe, and the quant picker — and restore them afterwards. Records, including DEBUG under drove -v, go to Textual's log instead; non-TUI commands and drove server log to stderr unchanged.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- cli: Require :QUANT for unattended multi-quant downloads (
2f0337e)
drove models download <repo> --yes on a repository with several GGUF quantizations now fails before touching the disk and asks for org/repo:QUANT instead of downloading every variant. Rerunning a download whose files all landed but whose record is still incomplete finishes the record instead of exiting early, models info reports an incomplete download with its reason, and the retry hint after an all-variants GGUF download keeps the picker. Consumer tests cover /v1/models, chat routing and agent registration.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- config: Name the variable for invalid members of JSON env values (
682946c)
An invalid element in an otherwise valid DROVE_ALLOWED_TOOLS list, or an invalid field in a DROVE_LLAMA_SERVER object, was reported as "DROVE_*" because any dotted error location was treated as a llama_server leaf. Attribution now starts from the root field and also considers the whole-field variable.
Findings: A:correctness-1, B:correctness-2
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- config: Report invalid paths as config errors, not tracebacks (
0a34382)
models_dir = 123 (TypeError) or a ~user path that cannot be expanded (RuntimeError) escaped the readable-error boundary because only ValueError becomes a pydantic validation error. The path validator now raises ValueError, so the file or DROVE_* variable and key are named and drove config <key> <value> can repair it. The same unexpandable ~ in --config or DROVE_CONFIG now gives a readable error too.
Finding: B:correctness-1
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- config: Save only the changed key and explain broken config (
0659bda)
Every config write started from the env-loaded Config and dumped all fields, so a one-off DROVE_* override (and every default) became permanent. Writers now patch one key in the raw TOML mapping and validate the merged file with an env-free settings class; init writes true defaults. Config.save is removed so no caller can reintroduce the bug.
Loading wraps TOML, validation and env-parse failures in ConfigError, naming the file or DROVE_* variable and the key; the CLI prints it and exits 1. config and init no longer load the config up front, so drove config <key> <value> can repair a broken key.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- downloader: Refuse non-GGUF repos and record download completion (
99770e9)
Text and vision downloads take GGUF files only; a repository with no GGUF or ONNX weights is refused with a message that drove does not convert weights. Every download now writes its record incomplete before the first transfer and complete after the last, so an interrupted download is never listed or served, and two repository files that would flatten onto one local path are refused before anything is written.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- model-store: Apply a symlinked primary's target records (correctness-1) (
e5f2f40)
A contained file symlink (alias/model.onnx or alias.gguf pointing into another model's directory) reached an unfinished or stale download without consulting its record, because records were looked up only along the lexical path. Also walk the resolved target's ancestry, and apply the split-shard rule to the target, at every call site. No-record links, a symlinked models directory and old downloads keep working.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- model-store: Honour every download record above a primary (correctness-1) ([
7b63781](https://github.com/cleanunicorn/drove/c...
v0.13.1
v0.13.1 (2026-10-27)
Bug Fixes
- cli: Reject unsafe model names in models commands and downloads (
ebded0b)
models delete, info and config validate the name through the shared rule and exit 1 with the rule spelled out, so delete . -y and delete .. -y no longer reach rmtree. download --name is checked before the models directory is created or the Hub is contacted, an explicit empty --name is no longer silently replaced by the inferred name, and DownloadPlan.destination validates the (mutable) local name and refuses a directory that resolves outside models_dir for every backend. add-tts and the TTS download path reuse the shared validator.
The containment helper becomes public (is_within) now that the downloader uses it. Download fixtures that used a name with a space switch to a hyphenated name, since whitespace is no longer valid.
Refs #110
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- downloader: Follow inward links when checking download targets (V2) (
ec1119c)
An inward symlink (for example .cache -> another folder in models_dir) passed the containment check, and the walk never looked behind it, so an outward link further down still let the Hub write outside models_dir. Check every target and everything under it by resolved real path, following directory links and tracking scanned real directories so link loops terminate.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- downloader: Refuse download targets that resolve outside models_dir (
146e0f5)
DownloadPlan only checked its destination directory, so a pre-existing symlink below it (a nested repository folder, a weight file, the Hub's .cache folder, or the sidecar config and TTS marker written afterwards) could redirect the download's reads and writes outside models_dir. Every one of those paths is now checked before local status is read or anything is downloaded; the CLI reports the offending path and exits 1, and the quant picker shows "local status unavailable" for such a plan.
Refs #110
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- downloader: Reject escaping symlinks anywhere under the Hub cache (V2) (
66d9582)
The previous check covered only dest/.cache itself, so an outward symlink at .cache/huggingface or a metadata/lock/partial file beneath it still let huggingface_hub write outside models_dir. Walk the existing cache subtree without following links and refuse the download if any entry escapes.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- model-store: List only model names that resolve accepts (
de0010e)
list() and complete() still offered on-disk names that break the new naming rule (a flat my model.gguf, a bad name/ directory or namespace), so a name copied from drove models list, completion or GET /v1/models failed with an invalid-name error. Inventory and completion now apply the same rule as resolve(); such models stay hidden until renamed on disk, as the docs and changelog now say. Completion also skips a flat GGUF whose sidecar escapes, matching list().
The model-list output test keeps its literal-markup check by putting the markup in the models directory path instead of a model name.
Refs #110
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- model-store: List, complete and resolve the same model candidate (correctness-4) (
f713ea6)
The sidecar fix made resolve() skip an unservable first candidate while list() dropped the whole model after checking only candidates[0], and complete() never checked servability at all. _add_entry now picks the first servable file in the same format priority as _find_primary, and complete() is derived from list() so the three always agree.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- model-store: Treat a model with an escaping sidecar as absent (
01a694c)
Config reads and writes follow a model's sidecar TOML, so a contained primary with a sidecar symlinked (or dangling) outside models_dir let models config, models info, the server and integrations read or overwrite an external file. The store now only serves, lists and resolves a primary whose sidecar also stays inside models_dir, and the HuggingFace alias scan skips TOML files that resolve outside it.
Deletion is unchanged: it unlinks a sidecar symlink without following it.
Refs #110
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- model-store: Validate model names and keep paths inside models_dir (
b7773a7)
Add validate_model_name, the one shared rule for model names (name or org/name safe-token segments, :QUANT allowed), and call it before any disk access in ModelStore.resolve and find_root. Every candidate path is also resolved and must stay strictly inside models_dir, so traversal names and symlinks that point outside the store are treated as absent; list() and complete() skip them too. Alias results that are not valid names are ignored.
The two CLI deletion tests that relied on an outward root or namespace symlink being a model now assert that deletion refuses it and leaves the link and its target untouched.
Refs #110
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- observe: Contain record files and unfiltered scans to observe_dir (
d0bcaaf)
save_record checked the model directory but then wrote through a symlink at the record file itself, and listing, search and lookup without a model filter followed directories and record files that point outside observe_dir. The record path is now checked before writing, and scans skip directories and symlinked record files that resolve outside observe_dir.
Refs #110
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- observe: Never write or read records outside observe_dir (
66a3f8b)
save_record validates a record's model through the shared rule (None still maps to _unknown) and refuses a model directory that resolves outside observe_dir, so a crafted model name or an outward namespace symlink writes nothing. A model filter on listing or lookup is treated as a key: an invalid or escaping one matches no records instead of reading another directory. The namespaced org/repo:QUANT layout is unchanged.
Refs #110
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- proxy: Reject an explicit null model with 400 (
6e9a3e6)
A JSON "model": null was treated like an omitted field and routed to the default model. It is a non-string model value, so it is now answered with HTTP 400 before ensure_running, like any other invalid model; only an omitted field falls back to the default or loaded model. The speech default-routing test's null case now expects the 400, and docs/api.md says so.
Refs #110
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- proxy: Reject an invalid model name with 400 before any lookup (
2ffad6a)
The proxy now tells an absent model field apart from a present one. Any present value that is not a valid model name (a non-string, an empty string, a traversal or absolute path, a file-valued multipart field) is answered with HTTP 400 before ensure_running, the model store or the disk is touched, and the request counters settle. JSON null and an omitted field still route to the default or loaded model; a configured default_model is validated the same way.
ServerManager.ensure_running validates on entry too, so non-canonical spellings such as ./qwen never become separate instance keys.
Refs #110
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- server: Keep backend output and shutdown from blocking (
e9aec63)
Backends were spawned with stdout piped but never read. Once ~192 KiB queued (kernel pipe plus asyncio's paused buffer), the next write blocked the worker; uvicorn's access log reached that in ~2,600 requests, freezing the ASR and TTS workers. stdout now goes to DEVNULL, stderr keeps its bounded drain, and both workers turn off the access log.
The wait after SIGKILL was unbounded and ran under the manager lock. asyncio only reports an exit once every pipe closes, so a descendant holding stderr hung every later lifecycle call. Bound that wait with STOP_TIMEOUT_SECONDS, warn on timeout, and always cancel the stderr reader.
Closes #114
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- server: Keep draining stderr after a stop gives up (
80b1bed)
Cancelling the stderr reader once the bounded post-kill wait expired left the pipe of a surviving descendant open with no consumer: the descendant blocked after ~64 KiB more output and the transport and fd leaked. Leave the reader running; it ends on EOF, after which asyncio releases the pipe and subprocess transports itself.
Co-Authored-By: Claude Opus 5.5 <n...
v0.13.0
v0.13.0 (2026-10-26)
Bug Fixes
- chat: Show a tool that crashes while running as failed (
ab087ba)
A tool that raised after it started (for example read_file with a null path) was relabelled "not run", which is reserved for calls a stop prevented. Finish the entry as failed with the error text, then let the existing error handling end the turn.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
Documentation
- chat: Document live tool feedback and the context meter (
3446ea5)
Describe the running tool entry and its outcomes, streamed shell output, the header context meter with its fallbacks and resets, and the new drove.context_length catalog field; add the Unreleased changelog entries.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
Features
- chat: Add a context-window meter to the header (
21bd876)
The chat now shows how full the model's context is after each reply: a bar-only meter at the top right with used / total tokens in its tooltip and on the Info screen, turning to the warning colour from 80%. The size comes from drove.context_length on /v1/models; when it is unknown the header shows the raw token count instead of a guessed fill, and with no usage the meter is hidden. Usage now belongs to the latest response, stays visible while the next reply streams, and resets on /new, model switch, resume, stop or error.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- chat: Advertise known context length on /v1/models (
f013c29)
The chat needs the active model's context size to draw a fill meter, but ModelSpec.context_window falls back to a guessed 4096. Report the size only when a sidecar, extra_args or GGUF metadata establishes it, as drove.context_length (null otherwise), and parse it strictly on the client.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- chat: Show running tools live with streamed shell output (
6714556)
A tool call now appears as soon as it starts, names the tool and its arguments, and updates in place to done, failed, denied or not run. The activity line names the running tool with elapsed seconds, and shell output streams into the entry through a bounded tail that a 10 Hz timer drains, so the worker thread never touches widgets. Stop keeps waiting for the tool and records its real result, as before.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
- tools: Stream shell output and mark failed tool results (
385764e)
The chat needs live shell output and a reliable failed/done distinction. Foreground shell_execute now drains stdout and stderr in raw chunks with one deadline and reports them to an optional callback; the result text sent to the model is unchanged. Errors, timeouts and nonzero exits come back as ToolFailure, a str subclass, so the UI never parses output to guess.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
Refactoring
- chat: Drop the unused tool-entry state tuple (
7875cfb)
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
Detailed Changes: v0.12.0...v0.13.0
v0.12.0
v0.12.0 (2026-10-25)
Bug Fixes
- chat: Save tip progress without persisting env overrides (
5ab02de)
Tip progress was saved via load_config()+Config.save(), which writes the fully resolved config — DROVE_* env overrides and every default — into the TOML file on the first Enter press. Persist only tui_completed_hints instead.
Also request stream_options.include_usage so llama-server actually reports token usage for Info, drop an unused "compact" CSS class, and fix CHANGELOG bullet spacing.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
Documentation
- Update installation and require temporary worktrees (
ecbb3ce)
Features
- chat: Reclaim conversation space with progressive guidance (
b7183bc)
Detailed Changes: v0.11.0...v0.12.0
v0.11.0
v0.10.0
v0.10.0 (2026-10-25)
Bug Fixes
- chat: Exclude unservable weights from model selection (
61b0ee3)
Resolve catalog entries through ModelStore before advertising chat capabilities. Keep unsupported weights visible in inventory with unknown capabilities.
Add six regression cases covering safetensors, bin and pt in local and namespaced directories, including availability after conversion to GGUF. All six failed before the fix; make fmt, make lint, make typecheck and make test pass (491 tests).
Sweep: 2 inventory-to-capability paths found; 1 fixed in the reviewed catalog change; 1 left in pre-existing bulk integration registration outside the review target.
Documentation
- Redesign documentation around user tasks (
7d84e00)
Features
- chat: Add searchable model selection and in-chat switching (
298c683)
Detailed Changes: v0.9.0...v0.10.0
v0.9.0
v0.9.0 (2026-10-24)
Bug Fixes
- stats: Capture audio usage in server status (
d3c3702)
Features
- tts: Add local KittenTTS 2 speech generation (
5690ea3)
Detailed Changes: v0.8.3...v0.9.0